For me, the official opcodes of the 6502 are enough to make a machine interesting and worth investigating, whether it's a TTL remake, an FPGA implementation, or an emulator.
But one could add more fidelity:
- some C02 opcodes
- some undefined opcodes
- accurate cycle counts
- cycle accurate bus behaviour
- cycle accurate interrupt behaviour
and all of these could be of interest, especially for intricately optimised game code or software protection code. But every one of those is an extra constraint, which makes the project more difficult to make, more difficult to verify, more complex and so perhaps more costly, and is likely to reduce clock speed.
I think choosing an appropriate degree of compatibility boils down to what a 6502 is for: is it a means of performing a computation, or a means of running a large body of existing software very accurately.
It might be the case, if you want to sell something, or maximise your audience, you might want to appeal to people who think of a home computer as being a games console. But I'm not at all sure why that audience might want something which runs at a high clock speed... so, it seems to me that the two goals, of high performance and a great degree of compatibility, wouldn't often be seen together.
(Not wanting to derail any specific threads or projects!)