Thanks for the thoughts.
Does a single prefix being withdrawn require a full recompute of BGP then? I know that the CCR1072 is slow on a full table - the mistake I made with a filter ended up pushing all prefixes from our ASR1k to the CCR, and even after I fixed it in less than a minute, it still spent a good 3-5 minutes adding and finally withdrawing the routes (about 250k at peak). The idea of using them as a reflector cluster was that once they'd loaded the table, they'd only be pushing the best routes to the core, and if one changed it wouldn't take it 5 minutes to send the update. I guess this then is not the case?
Could this be something to wait for ROS7, when multithread means more BGP power?
I might try the CHR on the core network as a route collector, and see how quickly it can build the full table.
CCR just from when it receives the announcement, until the route has been updated in the routing table can take as you say, 3 to 5 minutes - sometimes longer. Not only for the initial load, but for a path change too - especially when the BGP process (running on a single core) is busy due to a lot of processing (a lot of peers, full tables, etc.)
CHR does it within seconds (given a fast enough XEON or similar is available). Path changes happens pretty much on par to what you can except from a decent ASR or similar device. Initial full tables loads VERY fast (fastest I've ever seen to be honest).
Honestly, I wouldn't recommend that you waste your time with this on a CCR. Rather go CHR. I was also very hesitant initially with CHRs but they are proving to be rather stable, and much, much faster than the CCRs - in every aspect (not only BGP). If I could return my CCRs and get a refund for them, I would do so in a heartbeat. Even our cores and borders are being replaced with CHRs and we're taking CCRs out of the network, completely.
Also keep in mind, it's becoming rather common knowledge that the CCRs suffer from PSU failures too. Despite what MT claims, more and more users are coming forward with reports of PSU failures. The CCR IMHO, isn't ready for a critical service, such as what BGP is on the Enterprise network. It's definitely NOT something that you'd want to go down...