As it has been a while since the question was asked and no answers are present, I will try to shed some light. While the question is quite broad, I can hopefully give some context as to what might change the answer.
First, we'd need to define "leverage". Does that mean "sample telemetry to determine attitude"? Does it mean "use its control actuators (like reaction wheels)"?
Sampling Telemetry to Determine Attitude
If it's the former, the answer is somewhere in the neighborhood of "constantly", being anywhere from multiple times per second to once every X seconds, with X usually operator-defined, either through direct commands or through pre-defined modes in the ADCS system. The rate of telemetry collection would also greatly increase if attempting an attitude change using control actuators. You want to have high sampling rates especially to inform your fault detection, isolation, and recovery (FDIR) system to maintain systems and spacecraft health.
Using Control Actuators
If it's the latter, the answer is highly mission-dependent. You mentioned earth observation - is this a one-off earth observation satellite, one of a few in a small constellation, or one of an extensive commercial constellation, like that of Planet Labs? If you're a lone in a sun-synchronous orbit and want a picture of some area on Earth's surface at a certain time of daylight at a given exact moment, you may need to change attitude to get a sidelong look. In larger constellations, individual spacecraft have more well-defined areas of responsibility (so to speak) and so won't need to adjust attitude into or out of certain regimes as often.
You must also consider the size and type of data you're trying to download at a given ground station pass, how often the spacecraft comes into range of a ground station, the storage onboard the spacecraft, and the radios/antennae on the spacecraft. For example, if you're only download live and perhaps historical telemetry, you don't need to use a very directional, high-throughput radio to do that, and may not need to adjust your attitude all that much to get connection and downloads. You may also not need to take advantage of every ground station pass if you can download many hours of telemetry history in one pass.
If you're trying to download EO imagery, high-resolution images take quite a bit of storage space, and so your rate of imagery capture and amount of onboard storage will be the variables that determine how often you need to use communicate with ground stations. That communication that is likely to require a higher-throughput, more directional antenna to enable downloading greater amounts of data in those short periods of time. Using that more directional antenna will likely necessitate more precise attitude control.
Still talking EO imagery, if we're talking about visible-spectrum, RGB imagery, you're probably taking images only in daylight. That means that maybe half the time, you'll be essentially idle from a control actuator standpoint. So your question's answer would change depending on whether the spacecraft is in eclipse or not.
Attitude Maneuver Specificity
I'll address your question "Would it be employing its ADCS to optimize every potential ground station pass or target collect opportunity, or would it be a more rare occurrence[?]" more directly, still talking within the context of "it depends on the mission". How much you optimize ground station passes and target collect opportunities is often a function of what you need out of the pass or TCO and of the power available to you.
Hyper-precise pointing for a pass or TCO, combined with taking images themselves, can take a lot of energy. How much power is available to you, what power levels can you tolerate for spacecraft safety, how much power does a given action with given precision take? You gave a range for CubeSat dimensions of 6U to 27U. That is a huge range, relatively speaking, for potential solar power generation capacity, battery capacity, and available power.
To ask more pertinent questions for a TCO, what time of day do you want to take it, from what orbit, from what distance, at what resolution? What is your acceptable off-boresight angle for a captured image to be acceptable?
There's also the consideration of how rapidly you need to change spacecraft attitude, and how often you need to make these rapid attitude changes. If you need to do so very quickly, you might have reaction control thrusters (thrusters that impart rotational rather than linear motion) in your design, in addition to reaction wheels for your slow and steady changes. Your cadence of rapid attitude maneuvers will factor into how often these might be used.
You mentioned orbital transfers. These maneuvers are very costly, especially for CubeSats, which tend to have limited propellant capacity. In the past, some have not even had thrusters at all. CubeSats generally use their thrusters for station-keeping. How often the spacecraft must conduct station-keeping maneuvers is again influenced by mission parameters. Lower orbits will have a faster accumulation of atmospheric drag effects to counteract, and the relative perturbation effects by the Earth's oblateness and the gravity of the Sun and Moon vary with orbit as well.
If you're actually trying to conduct non-station-keeping orbital maneuvers in your mission, how often a thruster system is used for that will of course be determined by how often you conduct those maneuvers, and how accurate they are (that is, how much error there is to counteract in any subsequent correctional maneuvers you must conduct). Do note that such maneuvers will reduce your mission longevity, as you will not have as much propellant left for station-keeping to avoid reentry.
One of the advantages of CubeSats is how cheap they are. Dependent on available time and money and talent resources, it may be more advantageous to simply make more CubeSats to cover more orbital regimes than to try and move between them, especially if mission longevity is a large factor.
CubeSat Design Limitations
I mentioned earlier how some non-ADCS-design-related parts of CubeSats - their power generation, power capacity, etc - factor into the answer. Another design consideration that plays into this is the ADCS design itself. A key part of the ADCS in almost all CubeSats (and almost all satellites, for that matter) is reaction wheel/CMG desaturation. See Illmari Karonen's excellent explanation of how this all works.
Generally, at least in CubeSats in low Earth orbit (where my expertise lies), you're using magnetorquers to desaturate your reaction wheels. How often you need to turn them on is influenced by the control authority and speed of your reaction wheels and how quickly they reach saturation, along with how much control authority you get out of your magnetorquers (generally a matter of how strong of a field your magnetorquers can generate). It then follows that your electric power system design places limitations on these factors.
If you're using torque rods for your magnetorquers (quite common), you also generally need to be able to demagnetize the rods - that is, counteract the effects of magnetic hysteresis - which could itself be considered an ADCS control actuation system use. Furthermore, Earth's magnetic field's intensity is not uniform, and so the effectiveness of magnetorquers for desaturation and coarse attitude control (and therefore how often you want to use them) varies depending on the conditions in Earth's magnetic field where you are at a given moment.
I mentioned reaction control thrusters earlier - these can also be used for reaction wheel desaturation by exerting torque in opposite directions. This becomes more common in satellite design the further away you get from Earth's magnetic field, and is very common for deep space probes like Voyager 1 that will not have reliably-available strong magnetic fields to interact with using magnetorquers.
Fault Detection, Isolation, and Recovery
I mentioned in the sampling rate discussion the fault detection, isolation, and recovery (FDIR) system. In a nutshell (and as it applies to satellite systems and control engineering), the FDIR system monitors various internal and external telemetry to detect faults, diagnose their cause(s), and recover from those faults. Your recovery methods may themselves involve usage of control actuators. If so, the probability of a fault that requires control actuators for recovery, the bounds for what constitute a fault, and the control actuator use necessary for recovery all become factors as well in how often you use your control actuators.
To conclude, all of these different factors contribute to the very complex answer to the question "how often does a CubeSat leverage its ADCS". These factors can generally also be applied to answering the question as it applies to artificial Earth satellites in general.