I don't think there's a one-size-fits-all to answer this particular question. There are actually LOTS of variables to either approach - dyno or track. And both can be very good predictors of potential.
For many hardcore, successful racers the track is a perfectly good way to test their changes. They know their cars EXTREMELY well, they're consistent (car and drivers) and they have a lot of data as to how performance changes with their car at different tracks and under different conditions. Given all that -- they can very easily isolate very small performance changes for better or worse and know if they're headed in the right direction. And since, at the end of the day, success for them is ONLY measured by who got to the other end more quickly (8 seconds or 24 hours later) -- it's ONLY their performance at the track that matters. There are LOTS of cases where picking up a few HP at peak on the dyno didn't result in a quicker time at the track because other parts of the output curve were impacted as well -- in a negative way.
The other issue is that (and it's kind of unspoken in the thread) there's an implication that the kind of track racing we're talking about is drag racing. There are just as many autocrossers and road racers out there who do track testing too. And their performance envelope is also impacted by 2 items which the chassis dyno won't help you with - braking and cornering. In those areas - the dyno won't help at all. You've got to go to the track. And my experience has been that I gained WAY more in quicker lap times by optimizing my braking and handling than I ever did with HP additions. Beyond that -- as far as potential HP gains from mods -- it's REAL easy to see on the road course because you have long pulls in higher gears where traction is rarely an issue. If the braking and cornering speeds haven't changed - so you're exiting a turn at the same rpm in the same gear you always do -- but you're pulling 300 more rpm at the braking point for the next turn (end of the straight) -- then you've found some significant HP that's helping you on the track. In fact, you cant' use the same braking point -- as you find HP (more rpm is the measure) you have to have an earlier braking point because you're going faster. I tuned my road racer exclusively like that -- looking at rpm at the end of the longest straight -- traction not an issue. Driving skills were an issue though - you have to consistently take the turn leading ONTO the long straight.
The dyno can be a very helpful tuning tool. Almost all professional and successful racing outfits use engine and/or chassis dynos to help test things BEFORE they get to the track. And what they learn at the track is used to help them quantify what they saw on the dyno. In that way they use their on-track experience to help make their dyno runs even more predictive and predictable. But I think they'll also tell you that you have to be VERY careful to control variables on the dyno. I've seen 15-20 HP differences on chassis dyno runs on back to back pulls - the same combo with no engine changes at all; the only difference being changing air pressure in the rear tires, and tightening the tie down straps another click or two. The sad thing is that if you're trying to to test to find relatively minor changes you have to be very careful. Ambient temperature changes, oil temp/tranny and rearend fluid temps, minor mixture changes, tire pressure changes (which occur as the tires heat up from successive runs) can all result in measurable differences in output. When you're looking for 3 hp from a timing change, and you can have a 5 HP variation due to fluids heating up -- well, you see the problem. As a general rule of thumb -- I'd say HP comparisons are only valid for changes made during the same strap down session on the same dyno by the same operator with careful measurement of things (mostly temps) that impact output at the wheels. The minute you take it off and strap it back down -- you've changed things enough to NOT be able to definitely identify small output changes. Let alone comparing two different car's runs on different dynos at different times. That kind of comparison is closer to simple entertainment than it is to anything resembling an accurate or factual comparison of the 2 vehicles.
So, I'd say that both dyno and track have their place. Either can be predictive of potential -- but as with anything else - you have to know what you're doing. Having said that - I confess - I get just as much of a kick out of spending 3 or 4 hours at a 'dyno-day' watching different cars spin the rollers as I do at the drag strip.