What you described is a neat idea, but it's not possible with any degree of accuracy AFAIK. To give you a picture of the problem, calculating the disk usage of a directory requires calling statx(2) on every file in that directory, summing up the reported sizes, and then recursing into every subdirectory and starting over. The problem with doing a partial search is that all the data is at the leaves of the tree, so you'll miss some potentially very large files.
Picture if your program only traversed the first, say, three levels of subdirectories to get a rough estimate. If there was a 1TB file down another level, your program would miss it completely and get a very innaccurate estimate of the disk usage, so it wouldn't be useful at all for finding the biggest culprits. You have the same problem if you decide to stop counting after seeing N files, since file N+1 could be gigantic and you'd never know.