Thanks Russell-

I'm not actually using a data buffer, so there's no 'turnaround' time to take into account here.
Since the CF data bus is 'Hi Z' until sometime after the read strobe has been asserted, and the '552 data bus also 'Hi Z' at this time, I'd say that no actual 'hard' data collision could be taking place.

I can, however, see that the SanDisk devices do place data on the bus slightly quicker than the others (by 20ns or so), so maybe there's a still a residual amount of current on the data bus to be discharged at this time- I've got a few more ideas to try now (bus terminators, delaying read strobe) so thanks for your comments