In _Understanding the Apple //e_, Jim Sather writes that the NCR 65C02s
that he tried did not work, even though they were 2MHz chips. He makes
an educated guess that the problem is that data read from the main memory
on the ][ and ][+ arrives at the 65C02 with insufficient setup time (only about
50ns--technically within spec, but clearly marginal in the chips he tried).
The 6502 is much more forgiving, as are the Rockwell 65C02s.
Interestingly, he puts the GTE chips in the same category as the NCRs.
BTW, the data from the 16KB card arrives in plenty of time for all 65C02s.