Quoting Jason White (jason@jasonjgw.net):
These days, I simply have an MX record for my primary host; if it's unavailable, mail is queued by a secondary and I can just wait for delivery when the primary host returns, or run fetchmail -p etrn --fetchdomains <my-domain-name> <secondary-host> to hasten the process.
I stopped having backup MXes for two reasons:
1. People whom you rely upon to do backup MX for your domain proved to be flaky one time too often. (Back in Pleistocene days, it was common to serve as backup MX for each other as a courtesy.) Nothing quite as lovely as seeing a friend 55x-reject all of your domain's mail because he/she 'forgot' to preserve forwarding for your domain during his/her last rebuild.
And very hard to debug when you get occasional "hey I sent an email to you and it didn't go through". Years ago when I was just learning about this stuff we were having problems sending email to someone very occasionally and it turned out to be a rogue secondary.
2. Of course, that still leaves the options of backup MXing for yourself among multiple hosts of one's own, and of paying a commercial concern for the service.
Or having a mutual arrangement with a (trusted) 3rd party where you each run a virtual server at each others site.
In both cases, one hitch is spammers who preferentially dump spam onto high-numbered MXes (against RFC intent) on the plausible theory of MX secondaries tending to have looser antispam. The secondary MX then works hard to pump the relayed spam onto the primary, and various forms of badness can ensue including the primary working to teergrube the secondaries.
This "spam uses highest number MX" used to be a lot more common than it is now, to the point where you could exploit it by having a tertiary MX that always gave a "try again later" and the spambot would give up whilst having no impact on legitimate email. Additionally, I built some logic into my spam filter where mail could be rejected (or the spam score increased) by the secondary if the primary was known to be up. I don't see it happen nearly as often these days though, or maybe my filter is rejecting such things well enough that I don't see the recipient logged at the point of rejection. James