The upsides of migration basically come down to the following:
- (obviously) the reclamation of public IP addresses
- Transit interfaces are theoretically less directly reachable from the open Internet. I'd be really hesitant to call this a substantial security win, as you ought to be heavily filtering any traffic directly to or from your routers' interfaces.
The downsides tend to come down to a loss of visibility on the part of any external entity, including:
- Traceroutes will work but since reverse resolution of the transit IP's won't it will provide no insight to others as to what path a given packet takes (i.e. is 10.x.x.x a router in Ashburn, VA or Hong Kong? Who knows?). Worse yet it may show up as an internal resolution for some, which isn't a huge deal but may be confusing.
- Depending on your design, your network management systems ability to reach individual interfaces may suffer. You can potentially offset this by carrying those routes in your own tables with the (hopefully obvious) caveat that these prefixes need to be carefully managed and filtered to avoid exposing them to any third parties.
- Again, we don't know much about your network topology but if you're peering with others the extent to which you expose any of these transit routes (or need to) has to be analyzed. As an example, preserving your internal next-hop in a peering environment ceases to be a useful option.
- You may run into administrative issues with some providers - or other third parties - as "exposing" 1918 addresses to the broader Internet is considered bad form.
So - in practice with the information provided in the question it's hard to offer much more than the general observations above. I don't know how much space you'd potentially reclaim but would point out the additional (also hopefully obvious) point that the amount of work and potential disruption that migration will require is non-trivial: interruptions of service/change windows, changes to routing schemas and policies, monitoring tool adjustments, operational/troubleshooting protocols, etc.
Is the amount of time spent (or risk engendered) of greater or lesser value than the few thousand dollars the equivalent amount of public address space might be worth?