ksplice - interesting in theory, highly complex in practice
Sun Microsystems had a similar piece of technology (hotpatching / DUKS, aka 'dynamically update kernel services' or something like that) that shipped with early releases of Solaris 8. The facility was _removed_ it in later system updates.
Why is that ? Largely because of the complexity in managing a hotpatched system, and the complexity involved in creating such a hot patch for a given (likely already-patched) target system.
Patch management as-such is already complex - like, which order to install if multiple patches are applied, which services need to be started / stopped to apply certain patches safely, which patches can be applied without requiring a reboot to activate, which ones do need that. As your system ages, all these factors compound.
Now add splicing to the picture. At each of the above steps, for each of the patches involved, add the question "has it been spliced before", and the question "should it be spliced now", the question "what'd be the splicing baseline", "should it be spliced again after ordinary reboot, or patched ordinarily", and the same for "what if it crashes". Also, ask yourself the question how a splice/hotpatch can be qualified / verified / certified ...
In practice, this means splicing / hotpatching is a valuable tool for rare occurances only - namely, a well-understood, critical, but small and contained modification applied to _bridge_ the time till the next planned maintenance. To some, this is worth $$$$.
Most interesting bit about the news is that to Oracle, this must be an investment in an area where they already hold assets (the technology acquired with Sun Microsystems). If they are cooking up a good way of managing / reducing / eliminating the complexity mentioned, then I for one are looking forward to more. The hard part is not "how to splice", but rather how to manage / track a spliced deployment.
Although, of course, if you own enough patents on methods for "dynamically redirecting program flow", you can throw spanners into the gears of hypervisors, tracing utilities / mechanisms, virtual machine just-in-time compilers/optimizers, debuggers and all such. But then, these days any piece of software lives under the patent sword of damocles.