A soft fork is always better if possible because of the "danger of two competing blockchains"
There is no difference between the dangers of a soft fork and a hard fork.
In the event of a soft fork we have:
1.) The old chain exists with a more permissive set of rules.
2.) The new chain exists with a more restrictive set of rules.
In a hard fork we have:
1.) The old chain exists with a more restrictive set of rules.
2.) The new chain exists with a more permissive set of rules.
So they look exactly the same during a chain split.
The only difference is that a soft fork is backwards compatible because its more restrictive set of rules.
However, segwit expands the protocol (the definition of a hard fork), by using software kludges to make old nodes think they are compatible, and so is packaged as a soft fork.
Wouldn't segwit hard fork be better than soft fork?
Yes.
1.) We don't know how big a full block might be. It might be anywhere between 1MB and 4MB depending on usage. It is non deterministic. As a hard fork with 1:1 weighting this can be eliminated.
2.) There will be less technical debt by implementing segwit as a hard fork. The software kludges implementing it as a soft fork also creates huge maintenance risks in the future (segwit keys are 'anyonecanspend').
3.) All nodes will be able to perform their job properly. Segwit is not really backwards compatible, it's just an illusion created by the soft fork kludges feeding the old nodes filtered data and creates a two-tier node network.
Maybe they can be prohibited and all people owning bitcoins on non-Segwit keys have to transfer them to Segwit addresses.
I would consider it might be possible to make them eventually "send only" addresses after a grace period which allows users to change any known public receiving addresses.
I am against segwit as a soft fork. It should be a hard fork by its very nature and sets a dangerous precedent by using software kludges to package it as a soft fork in an illusionary manner. As a hard fork, it is possible that more could be achieved in this extensive protocol upgrade since it would not have design restrictions based on the imposed soft fork illusion.