2000.10.05 04:26 "Re: Varying BitsPerSample", by Sam Leffler
This was a conscious design decision based on my experiences--that the vast majority of usage had each sample the same width and that it wasn't worthwhile adding the complexity of varying bits per sample.
One other thing regarding this....
Every time someone would ask me to support something like this (varying bits/sample) we found other methods for handling their needs using facilities the library already supported. For example, the usual reason folks asked for vary bits/sample was that they had an alpha channel or similar that was different from the other channels. Rather than pack the data together it was better to just split the alpha data into a separate image/IFD within the same file. This allowed, for example, a different compression scheme to be used for each set of data. The only reason I can think of to have varying bits/sample is to support hardware codecs that can't split their data streams.
It's possible to support everything in the TIFF spec but I long ago decided that it was better for the library to handle only a "reasonable subset" and support that stuff as well as possible.