I saw it crash once after I let it sit for a couple of hours, but I haven't been able to reproduce that (of course).

I don't think that you're doing anything wrong. Certainly, the brush file looks fine.

The brush engine in FT is a variant of the Wilbur one, and the custom brushes come directly from there. I don't recall hearing about folks using the image brushes all that often, but it's quite possible that they have been having problems and just never reported it before. Is there any chance that a simple brush would be good enough for what you're trying to do? All of the brush engines use the same backend internally, but creating the image can be done in several ways: a built-in function (Simple), a user-specified function (Expression) or an explicit image (Image). There might be corruption in the Image front-end processor that is causing some problems, which is why I asked if Simple might be good enough.

What size is your editing data for the world? FT has an ugly habit of copying the whole surface during editing, which is why it tends to be laggier with larger editing resolution. Region-Of-Interest preservation has always been on my list of things to do for Wilbur and for FT, but I never really expected them to live long enough for the effort to have been worth it.

Personally, I haven't used brushes much since I implemented selections and the Mound tool because I find it much more controllable to drawn an outline and then add a mound based on that selection. I do admit that brushes can be useful (and fun) for prescale offset editing work, though.