First of all, cos 90° should return zero, and so does the 35s. On the other hand, cos 89° gives 0,0174524064373, the value you posted. And cos 89,99999° returns 1,7453292 E-7 both on your machine as well as on my early 35s bought in 2007.
So I do not think that anything has changed - the results of my early and your current 35s agree. The "bug" you mentioned refers to the loss of accuracy with arguments close to 90 degrees. In this case, the second result carries merely eight significant digits (out of twelve) with the last one truncated, not rounded:
cos 89,9999 1,74532925 E-6
cos 89,99999 1,7453292 E-7
cos 89,999999 1,745329 E-8
cos 89,9999999 1,74532 E-9
cos 89,99999999 1,74532925199 E-10 ...back again ;-)
Dieter