N.B. The email exchange below may have been edited, e.g. to remove content not essential to the main point(s) or to standardize English spelling/grammar.
I have been impressed with the materials that you provide on your site, and they have already been of use to me. I am getting into historical linguistics, particularly Indo-European historical phonology, after a time in other linguistic fields. One resource I have not been able to find, however, is the original data that the PIE reconstructions are based on. Are you aware of whether anyone has published the actual facts that Grimm et al. had in hand? I often see the same six or eight example cognate sets to illustrate Grimm's Law, for instance, but I have no sense whether the correspondence sets are based on six cognate sets or six hundred.
Thanks, A. B.
There are two books by Winfred P. Lehmann on our website that might help you. One of them is Lehmann's book Proto-Indo-European Phonology. The other is a collection of 19th century articles (incl. one by Jacob Grimm) that marked perhaps the first high tide in IE Historical Linguistics -- all of them in English. See A Reader in Nineteenth Century Historical Indo-European Linguistics.
As for your question, I don't know a precise number of cognates that were used, e.g. by Grimm. A fast, loose count of foreign language strings in the Grimm article in Lehmann's Reader totals 991. Probably not all of these are "cognate" citations, but the great bulk of them simply must be: starting roughly halfway thru that article, one notices long, long sequences of cognate citations. And of course these are only cognates cited in that single article by Grimm -- not by any means are these all of the cognates involved in his studies!
I'm sure the early IE linguists were working with sets of many hundreds -- quite possibly many thousands -- of cognates, but I just don't have a count available. Several other articles in Lehmann's Reader are notable for citing many, many cognates -- hundreds or, in Verner's article, multiple thousands of them.
You can always perform your own counts... I hope your command of Sanskrit, classical Greek & Latin, and ancient Germanic, Slavic, etc. is up to such a survey!
Re: more modern work, I looked at the second volume (the reflex index) of Julius Pokorny's Indogermanisches etymologisches Wörterbuch (1959), and I estimate that 75,000+ reflexes (old IE cognates) are listed therein: there are as many as 3*56 = 168 reflexes per index page, and there are 470 index pages in that volume, so arithmetic implies 78,960 reflexes; I discount that figure to allow for pages with fewer than the maximum 168 reflexes. Is 75K enough?
In summary, 19th century IE-ists were probably working with ca. 10-20 thousand cognates (offhand guess), but obviously could never include all of them in any article-length publication.
Welcome, J. S.