It's just profitable for them to pretend they're unable to process a lot of data during critical market events.
Really. Do tell.
Seriously, start thinking - stuff goes down because MtGox is the SOURCE of the information. It would affect any service that relies on the information supplied from the MtGox API, including Clark's site, btccharts, MtGoxLive itself, various bots, and more.
Bitcoin transactions to MtGox accounts can never be affected by them in terms of speed. Transactions FROM them however may be slower because they send them without fees, so slowness is expected for the first confirmation. However they would be slow any time, not just at certain times. Deepbit has a limit on the number of "free" transactions that they will process, and for a good reason too.
And as for bitcointalk going down, I don't know what caused the most recent downtime but to say it happens on every spike up or down is major BS. Even though it is hosted by Tibanne, that doesn't mean that they would just bring it down for some reason. Although it's just speculation, my personal guess is that the spike caused a virtual machine that runs the matching engine to balloon in memory size, shutting down other VMs in the process due to out-of-memory conditions. That is just a guess, though.
And as for the hardware itself, I don't know what they run currently, but there have been posts in the past that indicate a pair of high powered servers with around 96GB of RAM set up for failover. Not really small potatoes.