On pg. 255 fn. 22 I mention that Aulus Gellius (9.14) claims to have found 5th declension gen. sg. forms in -es in the manuscripts of Cicero (he quotes Sest. 28 equites vero daturos illius dies poenas) and—to state it more accurately than I did in the book—affirms his trust in those who claim that a manuscript in Vergil's own hand had dies as a gen. sg. at Georg. 1.208. At Sest. 28 the manuscript tradition does not apparently preserve any trace of the readings Gellius preferred, but at Georg. 1.208, according to the ap. crit. in the Hirtzel OCT, the manuscript known as π, the 9th century Codex Pragensis, is reported by Kvičala to preserve the reading dies.
There is at least one possible case of the 5th declension gen. sg. in -es found in the transmission of Cicero. Brent Vine points out to me that at Cael. 80 the reading plenam spes is found in B vs. the plenam spei of most mss. B is the name of the excerpts made by Bartolommeo Di Montepulciano from the lost manuscript of Cluny lent him by Poggio Bracciolini, which often preserves ancient readings.