I think Crosby's numbers actually show the potential limitations of this analysis for a single postseason. We have advanced stats for the 2016 playoffs, so we really don't need to use a goals-only analysis, and Crosby's 5-on-5 on-ice numbers were actually very strong (source:
Natural Stat Trick, so obviously using their definitions of scoring chances and high danger chances):
With Sidney Crosby on the ice in the 2016 playoffs (5-on-5):
Shots: 220 for, 159 against (58%)
Scoring chances: 216 for, 162 against (57%)
High danger chances: 85 for, 67 against (56%)
The problem was that his percentages were terrible:
On-ice shooting percentage: 7.27%
On-ice save percentage: .887
Goals: 16 for, 18 against (47%)
A full playoff season for a top line forward is still only about 400 or so total on-ice shots, and the normal variance on that can be insane, even between players on the same team. In the past three playoffs combined at 5-on-5, for example, Pittsburgh's save percentage is .899 with Crosby on the ice and .949 with Phil Kessel on the ice. The result is that Kessel is +18 at 5-on-5 compared to +10 for Crosby, even though Sid has 39 points at 5-on-5 compared to just 23 for Phil during that period (and despite Crosby having a better shot share, scoring chance share and high danger chance share as well).
Similar things likely happened to other elite players throughout history during specific playoff seasons, so I wouldn't be at all confident saying that a good or bad on-ice ratio necessarily reflects a player's actual two-way performance for a single playoffs, or even a couple of playoff seasons in a row. I think you need something on the order of at least 3-4 full regular seasons before you even start to get into something approaching an acceptable sample size for on-ice/off-ice goals numbers.