There are so many latencies in accessing memory, RAS, CAS, RAS to CAS and couple of others that I can't think of. In the worst case, truly random memory access all the latencies come into play. I think there are a max of something like 9 to 12 clock cycles (memory clock) of latencies. In the best case, fewer latencies are in play and CAS has a bigger impact. In any case, reducing CAS from 2.5 to 2 is only a half cycle and a small part in the total latency.
I've heard claims the net gains of CAS 2 vs CAS 2.5 is as low as 1%-2% or as high as 8%-10% for typical use. Personally I believe the lower estimate.
However, that said, in general CAS 2 memory is of higher quality than CAS 2.5 memory. You can often reduce the other latencies (RAS, tRCD, etc) of CAS 2 memory where you might not with CAS 2.5 memory (especially at high memory speeds). The more latencies you can reduce the better the performance.
<b>99% is great, unless you are talking about system stability</b>