For reference, Pierre from the dev team said that:
A branch mainnet-pack-validator has just been pushed to the gitlab repo tezos/tezos. This branch changes the way the chain is stored on the disk (dividing the size by an order of magnitude!). It requires, as a consequence to restart from a freshly imported snapshot (or to resync the chain from scratch (with --history-mode archive if you need it)). We are also not sure this is fully ready yet from critical infrastructures but this looks super promissing. If you launch the tezos-node with newly introduced option --multiprocess, in addition of the gain in size of the chain on disk, the "validation" of block is done by an independant process independently of the action of answering to the peers and the RPC of the node. This should imply a complete goodbye to freezes of the node... In one word, use with care but feelthe taste of the future (and report issues)!
Is it safe to use this branch for a production node with the purpose of reducing disk usage?