Yes it was a real repeatable problem for some, but it was sensitive to board layout as clock and serial pins are in close proximity in the 644 and 1284 dip packages. Some board layouts had the problem, others did not. First fix was a R/C filter on the rec pin which helped most seeing the problem but not all. Finally someone suggested changing the fuse bytes to full swing clock and that seemed to help anyone having the symptom. So in my 644/1284 boards I use full swing clock even though I've never seen the problem as Bob's board layout seemed to not bring out the symptom because of good board/trace layout.
I would try that 'fix', before blaming 3.3 Vs 5.0 voltage as I don't think anyone has demonstrated that causing a problem in a real world example, conservative datasheet specs recommendation aside.