In Ethereum's proposed proof-of-stake algorithm (POS):
A traditional node operator can't host a node themselves (e.g. on a home network), someone could just DDOS their node to bring them offline. This is only a problem in slashing (not POW) because it leads to a loss of funds (see later comment on this). So, a large percentage of nodes will likely be hosted on one of 3-5 cloud service providers. [These are the nodes needed for safe decentralization.]
Who would have the incentive to DDOS other nodes? Those with network resources, perhaps only those with the resources to have non-traditional DDOS protections on hardware outside these 3-5 cloud service providers. [These are multiple nodes controlled by one entity, like how Bitmain operates]
Are financial incentives everything? If yes, then this is a very real attack vector.
The owner of more resourceful nodes in #2 can just DDOS the ports of nodes hosted on one of these cloud service providers in #1. When network nodes drop offline for a time period, each node loses ~X% of their shares. It might be that the attacks are slow, such that only 0.5% of the network nodes are DDOSed offline per month. One prolonged DDOS attack could be enough to wipe out a node operator's yearly POS profits, and stop each of these operators from wanting to go back online. Over time, the coins of #2 become more valuable, and as more operators in #1 decide not to stake, so do #2's returns in proof-of-stake. So, the conclusion is that there is really a tremendous incentive for those with resources (#2) to DDOS nodes that actually decentralize the network (#1). I call this attack vector deliberate slashing.
In proof-of-work, the financial cost of custom ASICs is actually quite high. On the contrary, a DDOS attack on a POS node (deliberate slash) is cheap. Any POS algo with slashing included is at risk.
The part about nodes being incentivized to run on a cloud service provider is also significant. It substantially increases the insider threat risk for something to go wrong versus running on your own hardware. Because the risk of a deliberate slash doesn't require access to the OS, there are probably nuanced ways to affect a large number of nodes in this manner that may not exactly be considered a "DDOS attack". More concretely, it requires paying for markups on bandwidth costs that may not be there if you could safely run a POS node from home, giving those in #2 even more power.
Is it possible to answer this question without "Whataboutism"? Let's not get tribal and compare this only to other coins / consensus algorithms. Please stick to the engineering problem outlined here. The POS component of consensus is actually more centralized than the POW component because deliberate slashing is a much cheaper attack vector than building an ASIC -- but theoretically, just being more decentralized than POW wouldn't be enough anyway. If Ethereum switched to a POS algorithm without slashing, all these concerns wouldn't be valid.
I am not an expert in cryptocurrency or network stuff. But based on what little I know about peer-to-peer networks, this has to be something that was considered by the Ethereum powers-that-be. I.e., Joe Lubin's Infura and Etherscan.io. Is there a mechanism to prevent deliberate slashing?
EDIT: These concerns were actually brought up 12 months ago as well: https://www.reddit.com/r/ethereum/comments/7nkwzn/what_is_the_punishment_for_a_validator_node_being/ds36kei
Also, see this post regarding how offline nodes (like those that have been DDOS'd) may generate Uncles leading to a slashing condition: Casper slashes nodes with poor network connection?