Understanding Logging and Recovery in SQL Server
Replaying log records is called the REDO (or roll forward) phase of
recovery. Reverting log records is called the UNDO (or roll back)
phase of recovery. In other words, recovery will make sure that a
transaction and all its constituent log records are either redone or
undone.
Fast recovery allows a database to come online as soon as the REDO phase completes, before UNDO is run. This is made possible by lock logging, which records the locks that were applied by a transaction in the associated log record.
Fast recovery reads the log records that will be UNDONE, acquires the locks, then brings the database online. The UNDO processing can then proceed in due course while other database activity occurs as the resources that will be affected are protected by locks. See Paul Randal's Lock logging and fast recovery post for the gory details.
REDO operations are single threaded on Standard Edition and Enterprise Edition with less than 5 cores. On Enterprise with 5+ cores, there is a redo thread for every 4 cores (reference). This is separate and unrelated to the Fast Recovery feature as far as I can tell.
In answer to your other question, a single thread is responsible for transaction log writes.