Your first quote has nothing to do with the hosts file, per se. It involves the Opera internal check against its list of 'untrusted' sites supplied by Netcraft/PhishTank.

Your second quote (the link to the msdn blog) mentions a "bug" in Opera 10 that actually seems to make little sense as Jameson describes it, and one that I've never seen confirmed elsewhere. Browsers communicate with the Internet through the OS's network stack. That stack will use the target site's IP number if one is supplied by the browser, but if the stack receives a domain name instead, it will check the local hosts file and, if no such domain name is present there, will then check a DNS server to get the IP number. If a domain name is present (and correctly formatted) in the hosts file, the stack will direct the communication request to the location specified in the hosts file's entry line for that domain... often, that entry's target will be a "block" for the domain name, by directing such requests to 127.0.0.1. But none of that correlates with the author's few references that somehow that is all magically being bypassed by the 'bug' in Opera 10. And, presuming there actually were such a bug, it would have been long-ago fixed along the way, since there are no currently-open bugs on the various Cert-type software vulnerability listings.

Any browser that uses a proxy (including Opera's "Turbo" mode) will inherently bypass the local hosts file limitations by going out to the proxy server's address to make its connection requests, so the target site's domain-name lookups will be made at that proxy server, not through the user's local OS network stack and hosts file (other than to check the proxy's own address). If you are using a proxy, then your local hosts file will have no effect and will provide no blocking for browsed sites. Also, if the "blocking" entry in the hosts file is not correctly entered in the file, the lookup process will ignore that entry and will proceed as if the entry wasn't there.

I'm not aware of anything "odd" in KernelEx that would cause the hosts file to be bypassed by its mods... and it certainly never did for me when I last used KernelEx with Win 98 First Edition and Opera 10.52 a couple of years ago.