I've certainly seen some TIMs that are very soft and tend to squidge out, particularly on repeated use -- troubling for a prototype, but still fine in production. That includes some that incorporate fiberglass; I forget what material I'm thinking of, it might've been a Gap Pad(R), a gray one with just enough fiberglass to retain shape but it can crush down quite far if forced -- and it doesn't take much torque on a screw, with repeated application and the uneven pressure of TO-220s.
Note that, even if a TO-220 is biting in quite deeply, there can still be some fibers' thickness left, keeping them apart, and certainly not offering the full dielectric rating of virgin material, but probably still enough for functional purposes. And that can be fine for flat and smooth surfaces, but any minuscule protrusion, burr, etc. on either surface, and it's hard shorted. Which again in a prototyping context, you may have just errant metal filings floating around for example (even if doing cutting carefully, well away from the electronics; some can still get tracked in), that's not great. In production, that can be more carefully controlled, and it may also be hi-potted to confirm.
Note that the difference between functional and any kind of safety, is where the heatsink is grounded. If internal (insulated/shielded from outside contact; may be floating, or tied to local circuit ground), functional. If earthed, basic. If exposed to touch (and isolated, an SELV contact), reinforced.
Ed: as another, not really a case study, these are anecdotal, let's be fair -- I currently have some of this,
https://www.digikey.com/en/products/detail/bergquist/GPVOUS-0-020-01-0816/529937?s=N4IgTCBcDaIEYFMBOZUFoB2ATEBdAvkAwhich does a fine job in combination with even clamping pressure -- max clips like
https://www.mouser.com/ProductDetail/Aavid/MAX07NG?qs=%2Fha2pyFaduh3yz0bZfkSwKC%2F0csqKGJb%2FybLWvTVikPdiOobqWHfEw%3D%3Dare excellent.
Tim