Some of both, but I think the second effect is dominant. The rest of this post is all my analysis from scratch as a non-EE, so hopefully it makes sense.
Let's say we have an antenna at the end of a 100-meter (electrical length) lossless transmission line that's mismatched to the antenna by a 10:1 SWR. We'll assume that there's perfect reflection at the transmitter end of the feedline (so the transmitter is a perfect current source, doesn't dissipate anything, and can withstand whatever voltage is necessary).
10:1 VSWR is an 0.82 voltage reflection coefficient, ifwhich amounts to a 0.67 power reflection coefficient. If we send a pulse from the transmitter, 18%33% of power goes into the antenna and 82%67% of power gets reflected down the feedline, reflected again, and comes back for another try, 18%33% of that 82%67% is absorbed by the antenna, etc. etc. Those two trips along the feedline take 667 nanoseconds. $ 0.82^4 \approx 0.45 $$ 0.67^2 \approx 0.45 $, so after 31.5 round trips, over half of the power has made it into the antenna. Those 31.5 round trips take 2.33 microseconds1 microsecond, so any modulating signal with a bandwidth less than 430kHz1MHz will survive pretty much intact. $ 0.82^{23} \approx 0.01 $$ 0.67^{12} < 0.01 $, so after 2211.5 round-trips, more than 99% of the power has been delivered, and anything under 65kHz130kHz is going to have pretty much undetectable distortion. Since our signals on HF are almost always under 6kHz wide (and usually under 3kHz) this seems pretty safe.
AddingModifying our simplified system by adding loss anywhere in the systemany part of it, or having a more reasonable length formaking the mismatched elementsection shorter than 100m, or a bettermaking the SWR lower than 10:1, all make the reflections die down that much faster, giving a more favorable result.