The details are very sparse as to what they are actually doing. So, I'll speculate. Below are some claims and what I think they are doing.
The master key (in its plain unencrypted form) will be literally nowhere in the cloud, yet completely usable for split-key encryption.
Typically this is done using some sort of secret sharing, which doesn't necessarily mean homomorphic encryption (HE). That said, there are some advantages with using HE. Under typical secret sharing, the secret must be reconstructed at some point. With HE, that is no longer the case. Encryption and Decryption can be done homomorphically (see this and this). Timing results of both are promising and could be considered practical, especially for the first link.
we only have an encrypted form of the master keys. In more technical words, the Porticor product will perform split-key encryption with a partially homomorphically encrypted master key.
This leads me to believe they are doing something similar to the two links above. They store an encrypted version of the master key and their management key (similar to the safe deposit box they describe), and use those to encrypt/decrypt data. Most likely they are doing something similar to the first link as the result of that encryption/decryption is plaintext where the result of the 2nd link is not plaintext but the plaintext encrypted with the homomorphic key.
Assuming I am correct (who knows though as they don't publish the details yet), I would say it is practical. Look at the timing results in that paper. At the same time, it is somewhat marketing hype. Sure "The master key (in its plain unencrypted form) will be literally nowhere in the cloud, yet completely usable for split-key encryption" but the data must be decrypted at some point and can be stolen at that point in time. Also, I'm not sure I'd trust my data being protected by such new techniques yet. The field of homomorphic encryption is such an unstable field right now.