My memory was wrong. I found two papers. There may be an earlier paper with a different estimate, but these shown here roughly agree.
http://www.ira.cinvestav.mx/portals/0/Estudiantes/Tokuriki_2009.pdf
Experimental measurements in several different proteins indicate that the likelihood of mutation
to be deleterious is in the order of 33–40% [2,7,13] (36%,on average). Hence, as mutations accumulate, protein fitness declines exponentially [2]:W ≈ ea^0.36n (2)(where n is the average number of mutations) or even more than exponentially (see section on ‘epistatic effects’). So by the time an average protein accumulates, on average, five mutations, its fitness will decline to <20%. Thus, although the initial stability of a protein can buffer some of the destabilizing effects of mutations (Figure 1a), stability appears to comprise the main factor (although clearly not the only one [6]) that dictates the rate of protein evolution [1,4], and possibly of whole organisms [14,15,16], in particular, but not only, in relation to the acquisition of new functions.
Here is the figure I remember, but without numerical values.
Proteins can tolerate roughly 35-40% mutational substitutions. Note: this is not the same as comparing sequence differences between homologues, which have accumulated differences over time and stabilized them as they went. Rather this is the sudden introduction of mutations without time for stabilization. Here’s another.
https://jb.asm.org/content/jb/early/2016/01/28/JB.00831-15.full.pdf