I'll try to answer your immediate question first, then propose some alternate and more sensitive variations.
Let's start with a cubesat (maybe 3U) with a mass of 5kg and a miniaturized EM drive thruster (first peer-reviewed EM drive paper and pdf thereof) has been put inside, or maybe once in orbit the 3U "pops open" and a folded resonator made of a springy material with conductive inner coating expands to a nicer resonator size. One way or the other, let's assume the TM212 mode resonates properly with a high Q approaching something like the 7000 indicated in Figure 4 of that paper. That's pretty high, but let's just go with the flow. It does (potentially) violate basic physical laws anyway.
Ignoring the issue raised in @AndrewThompson 's comment about the thrust potentially scaling with the size of the cavity, let's use a simply scaled propulsive force for 5 Watts:
0.005 kW $\times$ 1 mN/kW = 5E-06 Newtons.
Since the force is so tiny, we should at least try to avoid as much atmospheric drag as possible. Let's choose an initial circular orbit with altitude of 800 km. Using an equatorial radius of 6378 km. the initial semi-major axis will be (in meters):
$$a_i = (6378 + 800) \ \times \ 1000$$
Using the vis-viva equation, setting r = a for a circular orbit and using a value of 3.986004418E+14 m^3/s^2 for the standard gravitational parameter of Earth $GM_e$, the initial orbital speed will be:
$$v_i = \sqrt{GM_e \ / \ a}$$
or about 7451.9 m/s.
Starting with some basic Newtonian physics:
$$ \Delta p = Force \times time$$
$$ \Delta v = Force \times time \ / \ mass$$
Five micro-Newtons for 30 days against a mass of 5kg gives a $\Delta v$ of 2.6 m/s, so yes your quick calculation is good.
Now a really interesting result. This illuminating answer by @MarkAdler confirms that, for a good rule-of-thumb for a very low tangential-only accelerations resulting in a gradual spiral between two co-planar circular orbits, the change in orbital speed will be negative the total $\Delta v$. So in the limit of zero thrust and infinite time, a total $\Delta v$ in the tangential direction of motion will actually slow the space craft's orbital speed by $\Delta v$ while simultaneously raising the altitude of the orbit!
So that gives a final orbital velocity:
$$v_f = v_i - \Delta v$$
or 7451.9 - 2.6 = 7449.3 m/s. Using the vis-viva equation again "in reverse" the final semi-major axis will be:
$$a_f = (GM_e / v_f)^2$$
or about 7183.0 km, an increase of 5 km. The period of a circular orbit is just the circumphrence divided by the speed:
$$ T = 2 \pi a / v$$,
so the period will change by 6 seconds, from about 6052 to 6058 seconds, which means the positions of cube sats in the two orbits would drift an additional 42 kilometers apart after every single 100 minute orbit!!
So actually, de-phasing will be a far more sensitive method than looking for an altitude change (as I've already mentioned here), and without considering all the other problems, it should be sensitive to $\Delta v$'s much lower than this.
So you can really turn the EM drive ON for one day, then off for one day, then repeat, and as long as you are tracking it by radar several times a day, some kind of test could be performed.
I am going to briefly suggest modifications to consider.
deploy a pair of matched cubesats, have them measure their relative positions by time-of-flight (light or radio) and have them alternate one OFF, the other ON every few hours or few days, depending on details.
give them GPS capability, but use a commercially available, orbit-qualified cubesat GPS unit, and not somebody's phone!
connect a pair together with a tether, or tether one to a dummy ballast, and (try to) get them to spin around their center of mass, or at least to deviate from their equilibrium position in orbit (the basis of an interesting follow-up question??)
