I've been looking at this article. This question is about the motion of fragments in the immediate aftermath of the collision. I've drawn the following figures and quotes from the link.
Figure 9 shows the evolution of the debris clouds 180 minutes post-collision, almost two revolutions later. The spread of each debris cloud around its respective orbit is already becoming apparent.
The next extract relates to a snapshot of about 30 seconds after the collision.
Examination of the interactive 3D scenario (provided below Figure 10) shows large out-of-plane relative velocity components, the apparent result of coupling of the two masses, despite the hypervelocity nature of the collision. There are also a large number of pieces of Cosmos 2251 debris with significant radial (downward) relative velocities, although it is not apparent why this situation would be the case. It is hoped that the availability of this data set will help researchers with expertise in hypervelocity impacts develop a more complete description of the collision geometry for this event.
This question is not about the mentioned radial (downward) velocities, though I grant it is curious.
The implication of the two screenshots, where figure 9 is three hours after figure 10, is that the spectrum of relative velocities is very small compared to original satellite velocities. The question is, why are there not more fragments immediately generated with a range of high velocity headings between the two original trajectories?
If it helps, think of the impact of two billiard-balls, in such a case neither ball retains its original heading and both depart in very different directions, according to the normal expectation of a near elastic collision.
I can see a possible explanation, that the collision was a glancing blow, the main bodies continued as they were and the spectrum of ejected fragments is near inelastic, in the sense that much collision energy is lost and so the relative velocities are low.
Does anyone have any more insight into this?