Somewhere I worked before we used to import 100,000s of a companies email into the project system to display them against the system. They were all in an Emails table which just stored the body, I think there was an email header table and an email contacts table which was relationally linked directly to the Person table.
Worked fantastically, used to run an IMAP server off those tables for outlook integration.
Per gmail account, you'd still dump all the mails into one gigantic emails table, either an SQL or a NoSQL one.
As for the totality of the gmail accounts, well, you'd use sharding, which is the same whether you use NoSQL or SQL.
You also want to cache the hell out of it, which is easier done in a NoSQL model (e.g. if the cache layer is integrated and hence uses the same API as the persistent stuff).
But a SQL DB can also use the same transparent API for the persistent stuff with their standard Query API. It wouldn't even need to parse the SQL statement, just hash it and compare it. And you could also use a layer that makes it transparent, such as an ORM.
I don't see anything in Gmail and Google Docs data usage needs that make NoSQL a better fit for them --only the need to scale with no sharing etc.
(OTOH Dropbox might fit the case, they only need some hierarchical store a la filesystem, with versioning).
the same way Word.exe doesn't use SQL internally, instead it uses linked lists and hashmaps
Yes, but Word.exe internally deals with just ONE document and it's data structures, does not manage metadata for many documents and does not have to provide aggregated info on them. That would be Windows Explorer, that manages many .docs. And that one, MS did have a plan to make more SQL-like.