It's never "recommended" to mix disk size's, but in this case (size is the only difference) you could do it even if you are adding them to a general pool. The issue you will need to forever monitor though is hot spots should your 900 GB disks fill up before your 1.8 TB disks do. Ultimately that will happen, if you plan to use the majority of that space.
If however, your space usage is low, but your extra spindles are for performance reasons, then it might never become an issue.

As I understand it, the 3PAR architecture is engineered as such to avoid these very problems that are more common in traditional arrays. The wide-striping/virtualization via the chunklet technology should balance the drive utilization across the system and reduce the proliferation of hot spots.

Consider the scenario; you have 20 total NL disks. 10 of them are 2TB disks and 10 of them are 4TB. In a balanced system, as data is written it is spread equally across all the disks. If the 2TB disks never get full, then there should be no penalty.

However, if you continue to write data such that the 2TB disks gets full, you will still have half the capacity remaining on the 4TB disks. So the system will continue to write to the 4TB disks. Now you have half the number of spindles handling your workload. When you read data, you now have an unbalanced workload because some reads benefit from all the spindles, and some from only half the spindles. The 4TB disks will have to work harder as they are returning more reads than the 2TB disks.

What you are saying for space consumption is true, but not for performance. You could put 2 blocks of data on a 4TB disk for every 1 block onto a 2TB disk and get even space utilization. But, you still have the 4TB disks having to work twice as hard as your 2TB disks because they are 2x more likely to have to read from their platter than the 2TB disks. The issue with that is the 4TB disks don't read/write/transfer data faster than the 2TB, it just stores more. Therefore, your overall performance could be impacted as you are hitting the larger disks with more I/O.

That's how I understand it also. It will write twice the amount of data to the 4TB than the 2TB, causing the 4TB to be loaded more until it is rebalanced based on hot cold data.