Here's my thinking on why there's no actual real goals on the scoreboard for them despite the great analytics. Individually, each of them brings positive analytics to the table. Each of them does certain things well. Taken in aggregate, as a "line," their individual positives add together for great "line" analytics.
Are they a line, though? Each of them plays a style that might not be conducive to the others. Each of them avoids doing certain "last-mile" type of things, as you point out, that can result in real goals. Why are Zib and Kreider highly effective off the rush, especially in odd-man situations, producing actual goals? There aren't many options on the rush other than to go forward and towards the net with less defenders to push back on them.
At 5v5, each of them can decide not to do a certain thing, like avoiding the wall or going to the slot and hoping a linemate will carry that part of the pact, Kreider flashes his speed off the rush. He occasionally flashes his size on the occasional cycle. Zib has pure straight-line speed that's effective on rushes. Zib doesn't do much of anything when the plays are of the cycling variety.
Their respective positive xG analytics has a lot to do with those 5v5 rush situations, as does their ability to take away the puck off stick checks to go the other way on the rush. Kakko's positive xG also comes from his ability to take the puck away from opponents and also from his ability to lug and protect the puck in the offensive zone. Goals are *expected* and vice versa for expected goals-against when these things occur.
But goals don't happen as much as they should because each of them plays a style that's not necessarily conducive to the others. They need to be able to really come together as a unit with that "last mile" stuff before they are working as a real "line." Shooting better would help tremendously also.
And I'll say this: Kreider being penalized nonstop for trucking over people cleanly early in his career really did a number on him psychologically as a physical player, turning him into a floating type of operator.