I can find their values and all details in the RFC, but I cannot find why it was decided to have this attribute in certificates, instead of leaving them to be multipurposes.
Is there a reasoning behind this choice (key length?) or is it strictly business related, as they are priced very differently?
Strictly speaking, a key should not be "multipurpose". Distinct key usages call for incompatible key life cycles. The Key Usage extension is a formalism of this fact.
For instance, keys which are used for signatures and authentication could be lost with relatively low consequences: if your smart card is destroyed, you can no longer sign, but no data is lost; you just need to be issued a new smart card (this is inconvenient, but no security breach occurred). But you do not want people to begin to encrypt data with that signature-only public key, because in that case you need to make backups of the key, to avoid data loss upon key loss -- the smart card will strongly oppose to backups, by construction. The Key Usage extension documents the idea that a signature key is to be used for signatures only because it won't be backupped. Conversely, an encryption key will normally be escrowed somewhere, and thus you do not want people to consider it as binding if it is used for signatures: this would allow whoever has access to the escrow/backup system to sign "in your name". There again, the Key Usage extension protects against that.
