There's no such linear relation between number of pixels and computation time. Noisier images take longer to process, as do those with ambiguities which require extra processing to resolve.
Unfortunately, until you finish, you're not going to know how much processing is required -- and by then it's too late for a progress bar.
If you wanted to put some effort into it, perhaps you could sample a bunch of images with different characteristics and see what the processing time correlates to extract a model for the processing time, but I have my doubts.
Tess has an internal pipeline which should give more and more accurate estimates about how much work there is to do, but you'd need to dig into it a little bit to see whether you could leverage it for your purposes.
Tom