All the other folks who have responded are wrong 
The "right" answer (as always) is "it depends". If you're familiar with a lot of mainframe environments, you've seen systems running at 97% on a pretty consistent basis. That's because the workloads are predictable and the management tools are very mature. We can't say the same thing for the typical ESX environment. The typical ESX workload is very spikey/variable and very environment specific. Some people can run their hosts at 85% CPU and 90% RAM with impunity; others may have to be much more conservative. Things to consider include: do you have cyclical workloads that have to be accounted for (i.e. monthly / quarterly / etc. tasks)? are you running in an HA enabled cluster? do you need the ability to surge quickly (like, for example, when Apple released the iPhone)? what is your lead time for provisioning additional capacity (if you're running at 80% and it takes you six weeks to stand up new capacity, is that OK?)? and others.
Also, is your site the failover site for a DR peer site? If so, then you need to make sure you have enough capacity to handle that contingency. So...as you see, there is no one right answer. You've got to look at your own particular environment and determine how much reserve capacity you need to meet your business requirements. Then you have to present those needs to your management team and let them weigh the risks against the costs and figure out how much reserve they're willing to fund, then you provision. Then, every six months or so, you review your infrastructure to see how well you're meeting your established guidelines and repeat the whole process again...
Ken Cline
Technical Director, Virtualization
TVAR Solutions, A Wells Landers Group Company
VMware Communities User Moderator
I was about to post something .... but Ken did beat me on time ... 
My view is that, as Ken suggested, this is not a mainframe. Mainframes regularly runs at 95/100% utilization (most of the time northbound) simply because what you get out of a mainframe is VERY predictable. Not even at which VM uses which resources .... but which VM needs to respond in this xyz amount of time. the x86s platform is light years far from this level of service (not implying it will never get there tough). On the mainframe you know that, if you touch parameter xyz your application is going to respond zyx% slower or faster (depending on the parameter).
I usually say that on the x86 platform you touch a bunch of parameters..... and you keep your fingers crossed. Well I guess I am exaggerating a bit but I think you know what I mean.
There is a reason by the way for which this is the case. The mainframe is more expensive than any x86 platform available so there are obviously motivations to run it close to 100% utilization. If you think for example at the x86 platform a "simple" upgrade from dual core CPUs to quad core CPUs would double your resources capacity basically at the same costs (both for the hardware and for the software). Yes we could ask VMware to invest into ESX at the point where you can drive workloads at 99% utilization and keep your %READY close to 0 (Vs raising to the roof)........... but then you couldn't complain if VMware is going to charge you 20K$ per socket for the hypervisor.....
Perhaps I digressed. Back to the question..... I would say 60/70% utilization is pretty much a number I would like to see out of my ESX host.
My 0.02 cents.
Massimo.
Thanks Ken and Massimo - you've saved me lots of typing
What I would suggest Jitendar is that, assuming you have no information on how 'spiky' your applications are, and reservng capacity for DR isn't a requirement, is to start at a low number - say 60% - and if after a month or two you find that the host has plenty of capacity that isn't being used by spikes (and backups/virus scans of the VMs can really spike their usage) then I would start to increase it. If that doesn't suit your virtualisation style then I would agree with Massimo and be conservative, as long as you are still achieving your company's goals for virtualisation especially the cost model (cost reduction compared to physical servers etc).
I'm flying Friday afternoon. My entire time will be here in FWB, so not so close to Orlando 
Heck, in your new role, I expect you'll be up in my neck of the woods on a not infrequent basis!
Ken Cline
Technical Director, Virtualization
TVAR Solutions, A Wells Landers Group Company
VMware Communities User Moderator