You might be surprised, because the answer is in the ballpark of 100 MHz when cost is no object. When cost is an object, well even simple designs like the MSI/LSI versions of the PDP-8 were clocked around a 1 microsecond memory cycle, comparable to most early microprocessors.
For an extreme example, the Cray-1 was constructed entirely from tens of thousands of discrete NOR gate chips using ECL, and was clocked at 80 MHz.
Multi-chip implementations remained standard on big systems into the 1980s. The ability to use bipolar technology over the much slower (at the time) MOS was a major factor, but it also simply wasn't possible to add to a single die sophisticated features like large register sets, pipelining, caches, etc. with the limited transistor budget available in the 1970s.
It wasn't until the mid-80s that CMOS microprocessors started to win decisively versus discrete logic for mid-range applications. It lasted even longer on the very high end -- DEC, IBM, Cray, and others, were manufacturing processors constructed out of many MSI/LSI chips well into the 1990s. Even the first POWER processor from IBM consisted of 9 VLSI chips, in 1990.
The overwhelming advantage for single-chip processors back then was the low cost. A minicomputer like the PDP-11/20 from all the way back in 1970 constructed out of SSI chips could run circles around the 6502, but the processor for the PDP-11/20 was like $2000 and the 6502 was $25.