- It required special expensive terminals, the VT330 (mono) or VT340. Those were not ubiquitous, the VT320 was the bread and butter type. Later VTx20 models also didn't gain support for it (usually the lower range models got premium features from the previous ranges like multiple connections / pages but this one didn't 'trickle down'). So there wasn't so much incentive for developers to support it as the installed base was very low.
- It required a long time to send all the information, because terminals were still on slow serial connections. The VT320 could only do up to 19200 bps which was not even super fast for text at 80x25! Sixel images would take ages to transfer. Even though we had different expectations of response speeds then, sixels were slow by the standard of the day.
- Colour highlighting (or even monochrome "bold"/ultrabright) was not even common in most commandline tools back then, even LS. Hell, on HP-UX in the 90s I couldn't even use the cursor keys in the default shell for command history and line editing! Mod cons we take for granted now were yet to be invented then.
- Thumbnail processing as the parent suggests would take a lot of space and processing speed. My quota at college was a few megabytes. In fact I hardly had any images. I think I had just one, a nice Enterprise-D render that I used as wallpaper. It's not like we had thousands of images to sift through like we do today. Of course the web changed that rapidly but at that time text terminals were already on the way out.
Some apps used it for stuff that was worth the wait. Like the print preview in word perfect. But many apps didn't bother because the above constraints just made them infeasible. Sixels only really took off after their original terminals were long gone. And really in this day and age we should be able to come up with something better.