Yes, I agree, sometimes they research really well, for example improving logistics so that they can get their fleets big enough to take on the human. If they get left to last they can easily master one of the weapon trees and one of the defence trees (to match your choice).
But sometimes they research all sorts of things that they don't need, like Warp Drive when they desperately need to cram weapons technology. It's not going to matter how fast their ships can move if they're going to explode without so much as denting the shields of an enemy ship. Personally, I only put engines on my ships if enemy ships are going to be fast enough to avoid interception, and when I do research propulsion, the standing speed bonuses that affect all ships are usually enough for me.
One of the factors holding back research is the continual upgrading of facilities. Every time you upgrade an existing building, the cost to upgrade increases. It isn't a huge amount but by the time you get all the way up to Discovery Spheres, you'd almost be better off demolishing and starting from scratch. I myself tend to turn off all automatic upgrades and do the upgrades manually when every second or third new building comes along. And sometimes I will demolish some of my existing structures so I'm actually producing the same amount of research on a planet but using fewer tiles to do it and paying less maintenance on buildings.
The AI really needs to get the idea that more buildings, especially ones which suck maintenance, is not always a good idea. Ultimately there are two things in its budget which will force it to set the industry slider to less than a hundred percent, and that is colony maintenance and ship maintenance. Ships at least can help it defeat other civs, or stave off attacks at least. Buildings are great big targets waiting to get washed away by Tidal Disruption.