I took a look at this, but it requires at least 8 lines segments to even vaguely approximate a circle or ellipse -- and double that to get somewhat close.

Even once several are combined, the inflection points created by compound curves crossing compound curves mean that there is very little scope for reduction in the total number of lines without loosing even the general shape.

And in the end, the point in polygon algorithm requires substantially more work/runtime than using one formula per ellipse; even when there is considerable overlap between them.

I knew it was a long shot from the outset; but I've been pleasantly surprised by such stabs in the dark before.