I think I've heard of some approaches to conceal the actual algorithm from the computing unit (by obfuscation, or by distributing the actual calculation between multiple units), given that the software we're actively using doesn't do anything secret - and is likely to be a free and open source software, it doesn't make much sense to protect the executable code besides integrity checks.
Otherwise, if we talk about - just for example - a receiving mail server (MDA) that has encrypted mailbox storage like [1], then it really depends on how much mail you receive. For a typical personal mailbox, where the steady rate is less than an email-per-minute (and usual email size is some kilobytes, not megabytes), encryption overhead is negligible. Such server would sleep most of time anyway.
[1]: https://grepular.com/Automatically_Encrypting_all_Incoming_E...