Note that there is a difference between routing logic and forwarding logic.
The latter is arguably simplified in IPv6; the former is essentially identical.
Variable Length Addresses were demonstrated by the TUBA team in 1994, with both Cisco and Proteon demonstrating slow and fast CPU paths and hardware assistance. The cost of handling fully variable lengths was noticeable, but vanished when a common length was chosen with uncommon lengths gated, rate-limited, quenched or otherwise controlled sourcewards.
In modern forwarding engine implementations using a dual between an m-way trie and associative real memories, the cost of a full VLA is now in the noise even for arbitrary streams of random-length VLA headers; the hard part is *still* the generation of the associative arrays from the routing tries. That is, the *routing* problem is the hard problem, not the forwarding. And VLAs can simplify the routing problem if they are designed with involuntary (proxy) aggregation in mind.
The early 1990s rejection of ideas from various IPNG proposals did not anticipate a mult-decade roll-out of the minimal changes settled on in SIP+PIP (which became IPv6), nor did it have any stubs whatsoever for adjusting the on-the-wire format in the future.
This exposes the biggest single problem with the ROAD/IPNG/IPv6 process: there was almost no thought in the working groups (which became increasingly detached from operators and middle-box vendors, and were dominated by systems vendors) to deployment scenarios that were very gradual and very local, with n-level enclaves of systems with just one protocol stack (e.g., an IPv6 only bubble inside an IPv4 only bubble attached to a the Internet via an IPv6 only gateway), and the hacks that have been developed to deal with such situations (which have arisen in real life) are at least as awkward as IPv4 NAT+address overloading.
IOW, it was all end-system-software-think and little to no thinking about broader issues on end systems (ones that are multiply attached to the rest of the world, notably, or ones that migrate from one network to another rapidly), and even less about routers (especially not routers that are themselves mobile).
The slogan, "every client is also a server" should have been extended to ".. and also a first-class router", which likely would have arrived at a better overall design for IPv6, and faster deployment.