Rails3: render_to_string of a .html.erb when request is a json request raises MissingTemplate error

I've the following situation (which actually works in Rails 2.3.8 and
this is a result of the migration of our app to rails3):

I have a partial which is rendered a few times (imagine tr's in a
table). On the clientside there's a script that does a json request
for a specific row (jQuery + ajax dataType: 'json' and type: 'POST')

In the controller I do:

render :json => { :success => true, :stuck_out_tongue: => render_to_string(:partial =>
'some_partial') }

Then the ajax result handler replaces the row with the data from the
json result.

This is 100% functional code in rails 2.3.8, but raises
MissingTemplate exception:

ActionView::MissingTemplate (Missing partial employees/
employee_list_row_data with
{:handlers=>[:haml, :builder, :erb, :rjs, :rhtml, :rxml], :formats=>[:json, :js, :"*/
*"], :locale=>[:en, :en]} in view paths ...

So it seems that the requested formats are used to search for the

First question: is this change deliberate? Is it now necessary to
have templates in the form of: "xxx.erb" (without the 'html') so it
can be used for requests in other formats? Removing the '.html' part
of the file fixes this problem. I just want to know whether this is as
expected or a bug.

Second question: when I look at the query (actionpack/lib/action_view/
template/resolver.rb line 71) it is like this: app/views/layouts/
{.haml,.builder,.erb,.rjs,.rhtml,.rxml,} You can see it uses the
request.format '*/*' also in the Dir[query]. Is this possibly a bug?
Because in (/home/cvp/myprojects/github/rails@github/actionpack/lib/
action_view/template/resolver.rb line 152 there's the following code

      # Overload formats= to reject [:"*/*"] values.
      def formats=(values)
        if values && values.size == 1
          value = values.first
          values = nil if value == :"*/*"
          values << :html if value == :js

I see there's specific code to handle '*/*' formats but only when this
is the only requested format. Is this possibly a bug?