Sebastien, ad 3. that is incorrect. Read the ticket more closely, look at the PR, look inside schema tool and platform classes. There is already a lot of quoting+unquoting being performed in 2.* and a lot of assumptions. Having quoting enabled across the board might actually increase performance in some cases, because there will be less scanning for keywords (see platform classes) and possibly less quoting/unquoting across Schema*.
The problem is, the quoting right now works in some places and in some platforms and is being performed only when schema/schematool/dql needs it, but is being ignored in all other cases. This means that columns like "group" or table names like "platform" will fail randomly depending on platform/rdbms you actually use. It's a nightmare with cross-platform apps and a struggle for single-platform apps, where your tables are named according to domain-rules and happen to overlap with some rdbms.
Quoting identifiers being "a bloat" is similar to saying, that implicit quoting values is a bloat. Although from security standpoint the former is much rarer, it's the same for portability and stability of the DBAL across platforms.